[mythtv-commits] Ticket #12045: Live TV after a short while always make the frontend hang

[MythTV]

#12045: Live TV after a short while always make the frontend hang
--------------------------------------+------------------------
     Reporter:  jyavenard             |      Owner:
         Type:  Bug Report - General  |     Status:  new
     Priority:  blocker               |  Milestone:  unknown
    Component:  MythTV - Recording    |    Version:  0.27-fixes
     Severity:  medium                |   Keywords:
Ticket locked:  0                     |
--------------------------------------+------------------------
 I'm very puzzled with this one.

 It only started since I've been back from a month away. I had never
 experienced it in the past.

 At first, I thought the issue with with my RAID array where the recordings
 are stored, as smart status of one of the disk started to be very verbose.
 Following that, I instead used as recording storage a NFS share to a very
 fast machine over gigabit (easily saturate the gigabit link and having
 local write speed in excess of 400MB/s on the NFS server)
 that didn't solve the problem.

 So I created a brand new backend, new disks, database on fast SSD (580MB/s
 read/write) the lot...
 still issue is just as bad, no matter the frontend.

 When starting live TV after a short while (ranging from 10-15s to 3-4
 minutes)
 the frontend will hang.

 In the backend log, you can see usually:
 {{{
 Feb  4 00:16:33 mediaserver-surrey1 mythbackend: mythbackend[27099]: E
 ProcessRequest ringbuffer.cpp:1105 (WaitForAvail)
 RingBuf(/external/recordings/1009_20140203131345.mpg): Waited 16 seconds
 for data, aborting.
 }}}

 I attempted to increase the various wait time, to know available.

 In fact, from the logs, it appears that once we enter a particular
 condition, we just never exit the waiting loop; it will always time out.

 in the logs usually there are plenty of:
 {{{
 Feb  4 00:15:41 mediaserver-surrey1 mythbackend: mythbackend[27099]: I
 ProcessRequest ringbuffer.cpp:1098 (WaitForAvail)
 RingBuf(/external/recordings/1009_20140203131345.mpg): Waited 0.5 seconds
 for data #012#011#011#011to become available... 182840 < 229376
 }}}

 which is fine, playback is just normal

 If suddenly the writer thread takes a little longer than usual for
 whatever reason, WaitForAvailable will start waiting forever:

 {{{
 Feb  4 00:15:53 mediaserver-surrey1 mythbackend: mythbackend[27099]: W
 TFWWrite threadedfilewriter.cpp:539 (DiskLoop)
 TFW(/external/recordings/1009_20140203131345.mpg:71): write(65424) cnt 22
 total 1362060 -- took a long time, 1425 ms
 Feb  4 00:15:53 mediaserver-surrey1 mythbackend: mythbackend[27099]: I
 ProcessRequest ringbuffer.cpp:1098 (WaitForAvail)
 RingBuf(/external/recordings/1009_20140203131345.mpg): Waited 8.0 seconds
 for data #012#011#011#011to become available... 137524 < 229376
 }}}

 To accentuate the issue and cause it to occur even more often, I run a
 side process, writing to the NFS share (not fast 10MB/s)

 There seems to be a racing condition somewhere, once the writer thread
 takes longer than usual once, then the reader thread will never recover.

 Attach is a log with the wait changed to 48s (WaitForAvail) and 20s
 (WaitForReadsAllowed).

--
Ticket URL: <https://code.mythtv.org/trac/ticket/12045>
MythTV <http://www.mythtv.org>
MythTV Media Center